I don’t like the AI writing style anymore. It’s very readable and it has great words, but it’s lacking imperfections. Like a raytraced 3D render of mathematically perfect shapes.
People forget that AI is trained on mediocre writing too, not everything a person writes is fire. Most of it is a mediocre, too long, and hard to understand; just like the outputs you get from LLMs.
It's ironically kinda less accessible in total, though. Because my browser lets me zoom in on a page almost infinitely, but I can only zoom out enough to make this text go from insanely-big to uncomfortably-large.
Depending on the company and their sector, people will nod in approval, or start laughing.
My company also accepted the risk of using Microsoft. We have a "data sharing agreement" together, with very powerful magical words. Compliance people are happy and sleep well.
For MCP servers, there's no need to install a potentially untrusted software on your computer. Remote MCP can do very little harm, a CLI though? You're vulnerable to bad actors or supply chain attacks.
It has a json schema, that’s the main point. It also enforces good documentation by design. No need to get a man page or run the help command, it’s in the context. It can work remotely with authentication.
Most CLI tools have JSON support. Your arguments fall flat pretty short.
I think MCP is fine in an env where you have no access to tools, but you cannot ripgrep your way through an MCP (unless you make an MCP that calls ripgrep on e.g. a repo, which in that case what are you doing).
Tool calls can have JSON schema enforced on lower level (token sampling). Although, I'm not sure if major providers do it, but I don't see any reason why they wouldn't.
I am not sure I agree on your statement about most CLI tools having JSON support. First, I’m not sure it’s true. But most are for sure not coming with JSON schemas for inputs and outputs.
GitHub Copilot Pro plus gives you access to many models and if you use them a lot, Microsoft loses money (though they can brag about one more active user and get your data).
It doesn’t give you the largest context windows, but most models perform poorly when the context is very big so it’s not a dealbreaker to me.
reply